Intrusion Detection System in Wireless Sensor Network using AI & ML

Abstract

The WSN systems can be severely threatened by cybersecurity attacks because of their distributed nature and resource limitations. In this study, a new framework for intrusion detection is proposed to perform efficient and effective cybersecurity attacks identification. Three different modules, which cooperate together, will be introduced in this study to provide better results in detecting attacks on WSNs. Swarm-Based Adaptive Feature Synthesizer (SAFS) is a module that creates an energy-efficient set of features by utilizing the concept of Particle Swarm Optimization to learn from the sensor behaviors and interactions. Also, Graph Attention Temporal Encoder (GATE) is another part of the presented model that exploits graph neural networks and attention mechanisms to consider the interactions between sensors and temporal traffic patterns as a key feature. Entropy-regularized Self-Supervised Detector (ERSSD) module detects the intrusion and abnormalities based on entropy regularization and self-supervised learning techniques in order to detect hidden threats in the network, without the necessity of a large volume of labeled training samples. In experiments performed in this paper, our approach achieved an accuracy rate of 95.3%. The presented IDS architecture is unique because it integrates the principles of swarm intelligence, graph-based modeling, and entropy-driven self-supervised learning into one pipeline that can be used for detecting intrusions in WSNs.